fix(export): onnx export issue

[ashwinvaidya17]

📝 Description

• Closes ❓ <title>Anomalib Studio setup #3543 and 🐞 Onnx Export: Onnx export fails for torch 2.9.0 and above #3050
• The problem is that the lowest torch version for python 3.14 is 2.9.
• However the default behaviour for torch.onnx.export from 2.9 is dynamo=True. So even when dynamo=None is passed, it defaults to True. One option is to pin to False (keep legacy) but imo we should update to the new default and slowly increase compatibility (https://docs.pytorch.org/tutorials/intermediate/torch_export_tutorial.html). They also renamed dynamic_axes to shapes
• This PR introduces warnings for legacy methods. It is possible that some of the models will not work with dynamo=True till we don't switch to functions like torch.cond

✨ Changes

Select what type of change your PR is:

• 🚀 New feature (non-breaking change which adds functionality)
• 🐞 Bug fix (non-breaking change which fixes an issue)
• 🔄 Refactor (non-breaking change which refactors the code base)
• ⚡ Performance improvements
• 🎨 Style changes (code style/formatting)
• 🧪 Tests (adding/modifying tests)
• 📚 Documentation update
• 📦 Build system changes
• 🚧 CI/CD configuration
• 🔧 Chore (general maintenance)
• 🔒 Security update
• 💥 Breaking change (fix or feature that would cause existing functionality to not work as expected)

✅ Checklist

Before you submit your pull request, please make sure you have completed the following steps:

• 📚 I have made the necessary updates to the documentation (if applicable).
• 🧪 I have written tests that support my changes and prove that my fix is effective or my feature works (if applicable).
• 🏷️ My PR title follows conventional commit format.

For more information about code review checklists, see the Code Review Checklist.

[Copilot]

Pull request overview

This PR updates anomalib’s ONNX/OpenVINO export path to better handle PyTorch 2.9+ behavior (notably the torch.onnx.export dynamo exporter), adds targeted regression tests, and relaxes PyTorch version constraints to unblock newer Python/PyTorch combinations.

Changes:

• Add export utility helpers to resolve dynamo defaults, translate dynamic_axes → dynamic_shapes, and raise an actionable error when onnxscript is missing.
• Update ExportMixin.to_onnx to support dynamo export while keeping the legacy exporter as default (with deprecation warning).
• Relax PyTorch upper bounds in optional dependency groups and add unit tests covering the new export behavior.

Reviewed changes

Copilot reviewed 6 out of 8 changed files in this pull request and generated 1 comment.

Show a summary per file

File	Description
src/anomalib/models/components/base/export_mixin/mixin.py	Switch ONNX export logic to explicitly control dynamo, handle dynamic_shapes, and centralize helpers.
src/anomalib/models/components/base/export_mixin/utils.py	New helper functions for export flag resolution, validation, dynamic-shape translation, and export-root creation.
src/anomalib/models/components/base/export_mixin/__init__.py	Adds package export for ExportMixin.
tests/unit/deploy/test_export_mixin.py	New unit tests asserting default legacy behavior, dynamo behavior, translation logic, and actionable onnxscript errors.
pyproject.toml	Remove torch<=2.8.0 caps for several extras to allow newer torch versions.
CHANGELOG.md	Document legacy exporter deprecation and dynamo-related export fixes under Unreleased.

Comments suppressed due to low confidence (2)

src/anomalib/models/components/base/export_mixin/mixin.py:1

• The copyright header was changed from a prior year to a single "2026" year. For modified files, the repo commonly uses a year range (e.g., src/anomalib/engine/engine.py:1 uses 2024-2026). Please update this header to reflect the original + current year range (e.g., 2024-2026) instead of only 2026.
  src/anomalib/models/components/base/export_mixin/mixin.py:131
• **kwargs: dict[str, Any] is not the correct typing for **kwargs. This annotation means each individual kwarg value must be a dict[str, Any], which is not what torch.onnx.export accepts and will confuse type-checkers. Please change this to **kwargs: Any (or a more specific set of explicit keyword parameters if you want stronger typing).

[Copilot]

validate_input_names only accepts list[str], but torch.onnx.export accepts any Sequence[str] for input_names. This is a breaking behavior change for callers that previously passed a tuple (or other sequence). Consider accepting Sequence[str] (and converting to list[str] internally) to keep the public API flexible while still validating element types.